self-hostedMITone Go binaryPostgreSQL 15+

Reliability
you can prove.

Define what reliable means for a service. cerbix measures it with checks from inside your own perimeter — and withholds any number the evidence cannot defend.

Get startedRead the docs
Try it — switch a probe off:
Window reportExample
checkout-api30d
99.953%
okcoverage 99.93%
Quoted against objective 99.9 · 46.5% of budget burned · rev 7.
ends at sealed_through, not now
17check types, with an SSRF guard on outbound probes
1 binarySPA, REST API and migrations embedded
5 rolesfrom a laptop to geo-distributed probers
0 bytesof your telemetry leave your network
Evidence before confidence

Most tools tell you a service is down. cerbix makes you say what reliable means — then holds you to it.

Every observation becomes GOOD, BAD or UNKNOWN with a reason. The window ends at the seal watermark, never at now. Below 95% decidable coverage the number is marked partial; with nothing to decide it is absent — never rounded to 100%.

How the SLI is computed →
42 min
A typical uptime check100.00%

1398 of 1,440 minutes produced a result, all up. The silence is simply not counted.

cerbix 100.000%
okcoverage 97.1%

42 min UNKNOWN — excluded from availability, counted against coverage.

Illustrative: one check, no failures — it simply stopped reporting.

01

Do not hide uncertainty

GOOD, BAD, UNKNOWN, no data, provisional and withheld are distinct states. Two independent coverage axes govern every aggregate.

GOOD · BAD · UNKNOWN(reason)
02

Measure inside your perimeter

Self-hosted. Probe private segments and remote regions from the inside — AMQP worker pools or an HTTPS-pull agent with no broker in the geo.

--role worker · --role agent
03

Turn evidence into action

The same Service drives SLO and error budget, incidents, status pages, on-call escalation and the release gate. One operating loop.

define → measure → respond
How it works

Every number is traceable to the rule that produced it.

Definitions are immutable revisions. You can always answer: under which definition was this month measured?

  1. 01 · Definerev 7

    Declare the Service

    Which checks are its SLI, how they aggregate, what is pageable. Click a member:

    all
    BAD
    any
    GOOD
    quorum 2 of 3
    UNKNOWN
  2. 02 · Observe

    Probe from the inside

    17 check types run from your perimeter, private segments and geo regions. cerbix produces its own observations.

    eu-centralus-eastdc-private
  3. 03 · Seal

    Reduce to one timeline

    GOOD / BAD / UNKNOWN, duration-weighted into one-minute buckets and sealed two minutes after each one closes.

    See the math →
  4. 04 · Act

    Drive the response

    SLO, error budget and burn rate; suppression, incidents, status pages, escalation and a release gate — from the same facts.

    incidentpagegate
What’s inside

One binary. The whole operating loop.

Read the introduction →
  • SLO, budget & burn rate24h, 7d, 30d, 90d — on sealed facts only.
  • Release gateALLOW / WARN / BLOCK / UNKNOWN for CI, with reasons.
  • Incidents & status pagesTimelines, postmortems, public service-first pages.
  • On-call that does not doubleRotations, ladders, suppression that fails open.
  • Geo probers, five rolesFrom one process to region pools and pull agents.
  • OIDC, RBAC, secrets at restMulti-tenant, 2FA, AES-256-GCM with rotation.
17 check types

Not a metrics backend. No query language, no log or trace ingestion. cerbix sits beside Prometheus, Grafana and PagerDuty — it answers whether a service is reliable by your definition. Where it fits →

See how evidence becomes a reliability decision.

Deploy your own in minutes: one binary and PostgreSQL.

Get startedRead the docsStar on GitHub